The neural network model was trained to perform object classification on the ILSVRC2012 ImageNet dataset (45), which contains around 1.2 million images. Each image was labeled with the category of the most prominent object depicted in the image. The dataset contains images of objects belonging to 1000 categories. The object classification accuracy of the model was evaluated on 50,000 images that were not seen by the model during training.

To examine its response to different numbers of items (i.e., numerosities), the network was presented with randomly generated images containing n = 1, 2, 4, 6, …30 dots. The network was tested under three different stimulus sets: a standard set and two control sets that controlled for non-numerical visual stimulus cues. In the standard condition, all the dots had about the same radius (standard set, r = 7 ± 0.7ϵ pixels, where ϵ was randomly drawn for a standard normal distribution separately for each dot). In the first control condition (control set 1), the total area of the dots and the average distance between pairs of dots were kept constant at 1200 pixels and 90 to 100 pixels, respectively. In the second control condition (control set 2), the convex hull of the dots was the same (a pentagon of constant circumference) regardless of numerosity (for numerosities larger than 4), and the shapes of the individual dots varied (possible shapes: circle, rectangle, ellipse, and triangle). The network’s responses were evaluated over n = 336 different images with an equal number of images (n = 7) for each numerosity and stimulus set combination. The sample sizes of total images and images of the same numerosity were adjusted to those applied in electrophysiological monkey experiments (20). For the numerosity matching task, the matching model was trained on a similarly generated but larger dataset of 4800 images and tested on a separate dataset of the same size.

Note: The content above has been extracted from a research article, so it may not display correctly.



Q&A
Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.



We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.